Question 2 of 6 Intermediate Scenario
How would you identify potential doublets and low-quality cells?
Short answer
Combine library-specific QC with doublet detection. Inspect genes and molecules per cell, mitochondrial fraction, expression markers and expected multiplet rate. A high doublet score is a warning, not automatic proof that a cell is a doublet.
Why
Quality control should consider each library separately because tissue types and cell populations differ in RNA content. Inspect distributions of detected genes, UMI counts, mitochondrial transcript fraction, cell cycle and obvious ambient RNA. Very few detected genes may indicate empty or damaged droplets, whereas unusually high counts can signal multiplets, but large activated cells can also be genuine.
Methods such as scDblFinder or Scrublet simulate or classify candidate doublets using expression and neighborhood structure. If a candidate cell carries both B-cell and T-cell marker programs, a doublet is plausible, yet true cell-cell interactions or unexpected biology cannot be ruled out from one score. Assess expected loading rate, coexpression of mutually exclusive lineage markers, cell hashing or genotype demultiplexing where available.
Avoid uniform hard cutoffs that discard metabolically active or rare populations. Document filtering and test whether inferred cell proportions change drastically under reasonable QC thresholds.
Common mistakes
- Deleting every high-UMI cell or treating algorithmic doublet scores as direct proof
The interviewer might ask next
What evidence would persuade you a high-RNA cluster is a genuine activated population?